As the title indicates, over multiple runs of my program, I get different values of trans and est. Are these local minimum? If so how do I get the optimal one?
NOTE - They all have the same starting value of trans and est.
Thanks!
Edit: I realized why I was getting different values of trans and est over different runs but it raised another question in my mind.
My objective is to train an HMM on classical song data (MFCC coefficients). To discretize it, I assigned the MFCC coefficients to 7 clusters.
I have a sequence of 10 observations stored in matrix M that I am providing to the hmmtrain function as input. Each row represents a song, and each column are the different clusters the MFCC coefficients (for different frames of the song) were assigned to.
ex:
M =
1 1 1 3 3 3 3 2 7 7 7
1 1 1 3 1 1 1 1 3 1 1
3 3 3 2 2 3 3 2 2 3 3
6 6 6 2 2 5 5 2 2 5 5
3 3 3 3 1 1 1 1 1 1 1
7 7 7 7 5 5 5 7 2 5 5
3 3 3 2 2 3 3 2 2 6 6
7 7 5 7 2 3 6 2 2 2 2
7 7 7 7 7 7 7 7 7 7 2
However, since I was recomputing the clustering every time, my input data would change. By saving the matrix to a file and using that every time my results were consistent. What I do not understand is why for certain formations of clusters (i.e. for certain input matrices) I would get better performance? Shouldn't cluster formation be the same by and large (even if the number ex. 1,2.. might represent a different cluster)?
Sorry about earlier. I'll repost the question if necessary. Thanks again!
